Cascade R-CNN: Delving into High Quality Object Detection

by Zhaowei Cai and Nuno Vasconcelos

This repository is written by Zhaowei Cai at UC San Diego, on the base of Detectron @ e8942c8.

Introduction

This repository re-implements Cascade R-CNN on the base of Detectron. Very consistent gains are available for all tested models, regardless of baseline strength.

This repository is still under construction. Will be updated very soon!

Citation

If you use our code/model/data, please cite our paper:

@inproceedings{cai18cascadercnn,
  author = {Zhaowei Cai and Nuno Vasconcelos},
  Title = {Cascade R-CNN: Delving into High Quality Object Detection},
  booktitle = {CVPR},
  Year  = {2018}
}

and Detectron:

@misc{Detectron2018,
  author =       {Ross Girshick and Ilija Radosavovic and Georgia Gkioxari and
                  Piotr Doll\'{a}r and Kaiming He},
  title =        {Detectron},
  howpublished = {\url{https://github.com/facebookresearch/detectron}},
  year =         {2018}
}

Benchmarking

End-to-End Faster & Mask R-CNN Baselines

^_backbone	^_type	^_lr schd	^_im/ gpu	^_box AP	^_box AP50	^_box AP75	^_mask AP	^_mask AP50	^_mask AP75	^{_{download links}}
^{_{R-50-FPN-baseline}}	^_Faster	^_1x	^₂	^_36.7	^_-	^_-	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{R-50-FPN-cascade}}	^_Faster	^_1x	^₂	^_40.9	^_59.0	^_44.6	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{R-101-FPN-baseline}}	^_Faster	^_1x	^₂	^_39.4	^_-	^_-	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{R-101-FPN-cascade}}	^_Faster	^_1x	^₂	^_42.8	^_61.4	^_46.1	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{X-101-64x4d-FPN-baseline}}	^_Faster	^_1x	^₁	^_41.5	^_-	^_-	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{X-101-64x4d-FPN-cascade}}	^_Faster	^_1x	^₁	^_45.4	^_64.0	^_49.8	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{X-101-32x8d-FPN-baseline}}	^_Faster	^_1x	^₁	^_41.3	^_-	^_-	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{X-101-32x8d-FPN-cascade}}	^_Faster	^_1x	^₁	^_44.7	^_63.7	^_48.8	^_-	^_-	^_-	^{_{model \| boxes}}
^{_{R-50-FPN-baseline}}	^_Mask	^_1x	^₂	^_37.7	^_-	^_-	^_33.9	^_-	^_-	^{_{model \| boxes \| masks}}
^{_{R-50-FPN-cascade}}	^_Mask	^_1x	^₂	^_41.3	^_59.6	^_44.9	^_35.4	^_56.2	^_37.8	^{_{model \| boxes \| masks}}
^{_{R-101-FPN-baseline}}	^_Mask	^_1x	^₂	^_40.0	^_-	^_-	^_35.9	^_-	^_-	^{_{model \| boxes \| masks}}
^{_{R-101-FPN-cascade}}	^_Mask	^_1x	^₂	^_-	^_-	^_-	^_-	^_-	^_-	^{_{model \| boxes \| masks}}
^{_{X-101-64x4d-FPN-baseline}}	^_Mask	^_1x	^₁	^_42.4	^_-	^_-	^_37.5	^_-	^_-	^{_{model \| boxes \| masks}}
^{_{X-101-64x4d-FPN-cascade}}	^_Mask	^_1x	^₁	^_45.9	^_64.4	^_50.2	^_38.8	^_61.3	^_41.6	^{_{model \| boxes \| masks}}
^{_{X-101-32x8d-FPN-baseline}}	^_Mask	^_1x	^₁	^_42.1	^_-	^_-	^_37.3	^_-	^_-	^{_{model \| boxes \| masks}}
^{_{X-101-32x8d-FPN-cascade}}	^_Mask	^_1x	^₁	^_45.8	^_64.1	^_50.3	^_38.6	^_60.6	^_41.5	^{_{model \| boxes \| masks}}

Mask R-CNN with Bells & Whistles